On Mon, 10 Jul 2017 11:53:31 -0300 Eduardo Habkost <ehabk...@redhat.com> wrote:
> On Mon, Jul 10, 2017 at 10:01:47AM +0200, Igor Mammedov wrote: > > On Fri, 7 Jul 2017 17:20:25 +0100 > > Mark Cave-Ayland <mark.cave-ayl...@ilande.co.uk> wrote: > > > > > On 07/07/17 16:07, Eduardo Habkost wrote: > > > > > > >> looks fine, > > > >> > > > >> so what I'd do is: > > > >> * drop 4/6 > > > > > > Yes. > > > > > > > Agreed on this point. But: > > > > > > > >> * make fw_cfg_find() use ambiguous argument and error_abort if > > > >> ambiguous == true > > > > > > During my latest tests I've found that everything works fine without the > > > ambiguous argument. > > > > > > Do we still want to keep it? And I don't think error_abort() is the > > > right thing to do here, I'd much rather return NULL and add a suitable > > > comment. > > I'd still use ambiguous argument and since you prefer not to assert > > I'd add errp argument to fw_cfg_find() and handle error at callsites. > > > > Just returning NULL isn't sufficient if you need to distinguish > > 'not found' vs 'duplicate' usecases, additionally 'not found' > > in most cases isn't even error but 'duplicate' definitely is. > > > > Aborting on diplicate in fw_cfg_find() is fine and would > > help to avoid touching current callers if you wish to limit > > patches scope, but you can go with proper error propagating > > route if you wish. > > Just making realize refuse to create two devices sounds much > simpler to me. No need to make fw_cfg_find() more complex (if we > add errp argument to it) or less useful (if we add > assert(!ambiguous) to it). the problem here was a error message to print if fw_cfg_find() returns NULL for missing or duplicate, if we need to print precise error we would need proper error handling. Considering to fw_cfg is builtin device I'd prefer just assert in fw_cfg_find() on duplicate (all the callers consider it as error) and let developer to deal with assert if it is triggered.